Artificial Intelligence vs Machine Learning
The only thing I find consistent about the definitions of Artificial Intelligence, Machine Learning and Predictive Analytics is there are many different overlapping definitions. I believe you can put a hundred experts on AI and ML in a room, ask them to define them and you will get 100 similar but different definitions. All will be willing to fight to the death that their definition is right.
I think of AI and ML and a few other terms like predictive and prescriptive analytics in the broad category of “augmented intelligence”. That is, independent of the category of technology or algorithms, the business objective is the same, to help us better understand the business situation by using the data to “figure things out”. Once it is determined that the algorithms are successful at "figuring things out" you can institutionalize them and automate the decision making or operational processes.
Augmented Intelligence Terminology and Examples
When applying “augmented intelligence” it is important to think about the questions you want it to answer. This will dictate the approach. I don’t want to oversimplify the methods to answer the question but I think terminology is important.
If you’re looking for a numeric answer like a sales forecast you could start with a category of AI called “supervised augmented intelligence” like regression algorithms. Regression is used when the output is a value.
If you're looking to drive categorization of a transaction or opportunity you could start with a category of AI called “supervised augmented intelligence” like a classification algorithm. An example is having AI provide an unbiased suggestion of the sales stage of a sales opportunity, such as “Sales Accepted Lead”, “Qualification”, etc.
Maybe you don’t know the question to ask but would like to learn more about how the business reacts to different actions reflected in the data. These type of questions are categorized as “unsupervised augmented intelligence”. If you want to understand the groupings of opportunities you would use “clustering” algorithms. If you want to understand the rules that describe large portions of your data you would use “association” algorithms. This could be used to describe the activities that separate “OK” sales reps from superstars so I can understand the characteristics of superstars and get all my sales reps performing like superstars. “Unsupervised augmented intelligence” is a tool to provide “intelligent insights” about my business.
The reason for calling out these categories of questions is:
To have an understanding that different algorithms are used to answer different types of questions*.
To provide the building blocks of AI, even though when you apply this to a business situation they all munge together. From a business perspective, you don’t care about the algorithms, from a design perspective, you do.
To understand AI is not a magic bullet where you throw in a bunch of data and outcomes insights to run with. Although software applications should insulate the user from all of these techniques.
In many cases, the answer provides less utility than understanding the drivers of the answer. As an example, we use regression analysis to provide a sales forecast** based on marketing and sales activity. A forecast is just a number. When I first start using it, I probably don't feel comfortable using the AI forecast as my target for the next quarter. What I would really like to know is what drives that number. What marketing programs and sales activity really work so I can understand the levers I have to drive sales.
As an example, we are going to look at a regression formula that forecasts sales based on marketing and sales activities. In the XPexample, X1, X2, Xp represent the quantity of different marketing and sales activities the sales leads/opportunities have participated in. b1, b2, bp represent the impact each of those activities has on the sale process.
Although the formula produces a sales forecast[Y], just as important as b1, b2, because that determines how much impact a marketing or sales activity has on the sales forecast. If b1, b2,...bp is zero or close to zero it means it has little impact on Sales (Y). You could have customers/prospects with 100’s of those activities and it doesn’t move the sales needle. Likewise, if b1 is a large number, it says X1 has a big impact on the sale and you would want to maximize X1 to drive overall sales.
AI Should Not Be A Black Box
Marketing hype tends to portray AI as a magic bullet, plug it in, turn it on and let it do everything for you. Actual implementations show the less black box it is, the more it is used for augmented intelligence, and the more utility it provides the customer. The utility is created by uncovering the key drivers and managing them. The following are two examples:
This first example shows how AI can augment decisions. In this example, the AI calculated sales forecast*** is set alongside manual and system aggregations for self-reporting, management reporting, quota, and pipeline to provide an unbiased perspective of the direction of sales and what the system thinks can be accomplished.
The business value of AI can be enhanced by providing drill down and reduce the Black Box nature of the forecast. The above example shows the sales team under William Burris. Drill down allows me to interrogate the results at the sales rep and opportunity level and help “confirm” or “deny” the AI forecast and feel more (or less) comfortable in it. If you don't feel comfortable with it, there are also levers to pull to change the AI forecast to "war game" key drivers. This helps understand the pattern AI sees in the data and sensitivities to the forecast.
The second example shows the sales opportunity health score. It includes the score and the key activities and characteristics that drive the score. Rather than just knowing that the score is high or low, you understand what the system see’s as important and the activities required to increase the health score. The individual calculations that drive the health score are also used to create alerts and can be used as a basis for prescribing the next steps for this opportunity. In this example, all values add, but you can also have activities that are detractors of the health score (no customer interaction in the last two weeks could be a detractor). Although not shown here, time is also a consideration in these calculations. Activities three months ago may have little impact on the current score.
Key Ingredient for using AI Successfully
A key ingredient for using AI is having lots of observations. In the sales process, this means collecting as many interactions between the buyer and the seller as reasonably possible. If sales reps provide little information about their interactions with the prospect it will be, hard to apply AI in a meaningful way to provide accurate forecasts, next steps, judge pipeline health, etc. In a subsequent blog we will review AI for helping collect interactions.
A second key ingredient is a time. Positive activity (emails, positive phone calls, web visits) that happened this week is a good thing (probably). If that same activity was a month or two ago and nothing has happened since it probably signals a problem with the account as it feels like they are less interested. Algorithms have to have a time component.
Watch the video https://www.youtube.com/watch?v=-oDbVGficHw